Anthropic AI Discovered New Bacteriophage Enzyme System
Researchers identified array-associated reverse transcriptases using LLMs to scan protein clusters.
Updated on Sept. 23, 2026 in Biotech
Live Poll
Do you trust AI agents to conduct original scientific research without direct human oversight?
Anthropic researchers have used the Claude AI model to identify a previously undescribed enzyme system within bacteriophage DNA. The findings, published in a company preprint, were generated after the model analyzed 1.9 billion protein clusters.
Why it matters
This research demonstrates the increasing utility of large language models in accelerating biological discovery by automating the identification of novel genetic patterns. It signals a shift toward AI-driven mining of massive protein databases to uncover functional elements that were previously overlooked by traditional screening methods.
The discovery process utilized Claude Mythos 5 to scan 1.9 billion protein clusters across 21.5 hours and 215.6 million tokens. The identified ART arrays consist of an enzyme, a partner gene, and 3 to 21 copies of a short DNA repeat, notably lacking the cas genes found in typical CRISPR systems.
The players
Anthropic
An AI research lab focused on building safe and capable foundation models, recently expanding its stack into biological discovery and molecular biology.
Claude
Anthropic's flagship family of large language models capable of complex reasoning and data analysis across scientific domains.
Coefficient Bio
A startup acquired by Anthropic in April 2026 to bolster its internal capabilities in synthetic biology and protein engineering.
OpenAI
A leading AI research organization that launched the life sciences model GPT-Rosalind in April 2026.
The details
The research team tasked the AI with identifying reverse transcriptase (RT) systems—enzymes that convert RNA into DNA—within bacteriophage, viruses that infect bacteria. By processing 949 agent sessions, the model narrowed 200,000 recovered clusters down to 3,564 candidate families before human reviewers analyzed 19 final reports. This approach effectively uses language model reasoning to correlate protein sequences with flanking DNA repeats, bypassing the need for manual library screening in identifying novel genomic architecture.
Timeline
Spring 2026: Anthropic established its new molecular biology group.
April 2026: Anthropic acquired Coefficient Bio and OpenAI launched GPT-Rosalind.
June 2026: Anthropic launched Claude Code and Claude Science.
September 23, 2026: Anthropic announced the discovery in a public blog post.
The Tech Race
This development highlights an intensified research competition between AI labs aiming to automate complex biological hypothesis generation. It follows the industry trend set by the launch of OpenAI's GPT-Rosalind, as companies increasingly deploy proprietary models to identify novel genetic systems.
While this discovery remains in the research stage, it establishes a workflow for mining protein data that could eventually accelerate drug and synthetic biology developments. Anthropic is currently seeking research proposals from external scientists to further investigate the function of these enzyme systems.
The takeaway
Anthropic's identification of the ART system validates the use of LLMs to parse massive biological datasets for novel genetic architecture. Readers should monitor the company's progress on their experimental determination of the enzyme's function and the results of their open research proposal campaign.
Further reading
For more on how AI is reshaping experimental pipelines, visit Biotech.
Live Poll
Do you trust AI agents to conduct original scientific research without direct human oversight?
